US-based Ambarella, a long-time supplier of camera chips, is betting that artificial intelligence will increasingly run on devices rather than in the Cloud.
Speaking to IoT Insider at embedded world 2026, Jerome Gigot, VP of Marketing, Edge AI at Ambarella, outlined the company’s strategy, which has evolved over 21 years from broadcast encoders and low-power video chips to AI-enabled sensors in security, robotics, and industrial applications.
Ambarella’s proprietary processor, CVflow, combines computer vision and perception software for enterprise and consumer devices. The company claims to have shipped more than 400 million system-on-chips, including over 42 million AI-enabled SOCs embedded in cameras, robots, and video conferencing systems.
“The big difference between us and GPUs is that we are algorithm first,” Gigot said. “We optimise for real-world perception tasks rather than general-purpose computing.”
Ambarella defines “physical AI” as artificial intelligence that interacts with the real world via cameras, LiDAR, radar, or microphones. Security cameras remain its largest market, but the company is expanding into retail, logistics, and manufacturing, where AI is used for everything from shoplifting detection and queue monitoring to defect detection on assembly lines.
Edge computing is central to Ambarella’s approach. Processing data locally, on the device or a nearby Edge box, reduces latency, lowers cost, and improves security compared with cloud-based solutions. “Edge AI is more efficient, secure, and cost-effective than Cloud processing,” said Gigot. Ambarella’s hybrid model combines on-device AI for immediate tasks, Edge boxes for medium workloads, and Cloud computing for larger-scale analysis.
According to the company’s latest financial results, 80% of its revenue now stems from Edge AI, reflecting a 50% growth in this segment last year.
Ambarella is also opening its platform to developers. Its newly launched Developer Zone offers SDKs, pre-trained models, and low-code/no-code tools, enabling third-party software vendors to build vertical-specific solutions. Gigot described the company’s approach as a “chessboard,” combining chips, cameras, and AI boxes to tackle problems across sectors.
Looking ahead, Ambarella is developing a next-generation chip using a 3-nanometre process, promising higher AI performance at lower power, critical for devices operating at the Edge. The company continues to expand its camera product line while growing its software ecosystem.
The company insists that privacy and ethics remain a consideration. Ambarella chips include features such as face-blurring masks, and the company restricts sales in certain countries. Nonetheless, the company admits that, like all tool makers, it has little control over how its products are used once sold.
“We’ve been doing physical AI for 20 years,” Gigot said. “Others are moving from Cloud to Edge. We started from the real world and are building upwards.”
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by visiting our LinkedIn page.
